Integrating deep time series prediction and intelligent control for optimization of tobacco drying process

In the tobacco industry, traditional moisture control for thin-plate drying equipment primarily relies on operator experience combined with proportional–integral–derivative (PID) control to adjust cylinder wall temperature. This mode suffers from heavy dependence on manual intervention and difficulty in achieving stable automated regulation of tobacco moisture content. To solve this problem, this study establishes an integrated intelligent closed-loop control framework combining convolutional neural network, long short-term memory network, and attention mechanism. The framework first builds a time series prediction model to learn process characteristics and predict the ideal reference cylinder wall temperature; it then dynamically corrects the temperature setpoint through moisture feedback, forming a supervisory control architecture embedded in the original PID loop. The prediction model reduces root mean square error by at least 33.05%, 23.17%, and 27.44% for the three tobacco brands, respectively, compared with baseline models, showing high fitting and forecasting accuracy. Experimental validation on a real cigarette production line verifies that the proposed method achieves precise and stable regulation of outlet moisture content, significantly reducing fluctuations around the target value. This approach greatly reduces manual parameter tuning while guaranteeing product quality, offering a practical intelligent regulation solution for tobacco drying processes.